Enclosure-Based Product Image Annotation for AI Detection Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Retailers face significant time and resource challenges in developing and constantly retraining AI models to detect a large number of products in their stores, especially due to frequent changes in product offerings and packaging, which affects the accuracy of product detection.
Innovation Solution
A centralized product detection system that utilizes a product detector trained by multiple product sources to automatically update AI models with new products and packaging changes, allowing retailers to subscribe to this system for accurate product detection without maintaining their own models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If retailers develop their own AI models to detect products in their stores, then product detection accuracy is improved, but time and resource consumption increases significantly
Solution Approach 1:
The patent introduces a centralized product detector service as an intermediary between product sources and retailers. This service collects images from multiple product sources, trains AI models centrally, and provides detection capabilities to retailers without requiring them to develop their own models. The service acts as a mediator that handles the complex model training and maintenance tasks.
Solution Approach 2:
The patent implements a centralized system that creates and maintains copies of AI models that can be deployed to multiple retailers. Instead of each retailer developing unique models, the centralized service creates model copies that can be distributed and updated uniformly across all client retailers, reducing redundant development efforts.
2Reliability
If retailers constantly retrain AI models to detect new products and packaging changes, then detection reliability is improved, but productivity decreases due to continuous retraining requirements
Solution Approach 1:
The patent implements preliminary action by having product sources submit product images and metadata in advance to the centralized product detector. The system proactively trains models with new products before they reach retailers, so that when retailers deploy the models, the detection capability is already updated and ready, eliminating the need for retailers to perform retraining operations.
Solution Approach 2:
The system implements feedback mechanisms where the centralized product detector continuously receives information about new products and packaging changes from product sources, retrains models accordingly, and distributes updated models back to retailers. This closed-loop feedback ensures detection reliability is maintained without requiring manual intervention from retailers.
3Adaptability or versatility
If retailers maintain their own AI models, then adaptability to specific store needs is improved, but device complexity increases
Solution Approach 1:
The centralized product detector service acts as an intermediary that handles the complexity of model maintenance, updates, and coordination. Retailers can access detection capabilities through this service without needing to manage the underlying model complexity, reducing their device complexity while maintaining adaptability through the service's ability to provide updated models.
4Measurement precision
If AI models are trained with comprehensive product data from multiple sources, then detection precision is improved, but the quantity of data and processing requirements increase
Solution Approach 1:
The patent merges product image data from multiple product sources into a centralized training system. By combining data from various sources under one coordinated training process, the system achieves comprehensive product coverage and improved detection precision without requiring each individual retailer to collect and process large volumes of data independently.
Data Source
AI summary
A method of obtaining images and data to train an AI model for product detection includes generating, with an image collection system at a first product source, a first annotation package including one or more images of and data about a first product. The first product is placed in a first enclosure located at the first product source. A process for obtaining the one or more images of the first product is initiated in the first enclosure. The one or more images of the first product is obtained with one or more cameras positioned in the first enclosure. The one or more images of the first product is provided to an edge compute device. Data about the first product is input, using an input device, into the edge compute device. An annotation package for the first product is created that includes the one or more images and the data.